What problem does it solve? ai-badger's hooks run silently, so questions like "did that hook even fire?" or "why is the drift notice silent?" have no answer without adding print statements and re-scaffolding. This Skill turns the framework's own behavior into a queryable record, making silent hooks, stale plugin copies, and version skew visible. ## Core Features & Use Cases - Toggleable audit logging: Enable hook logging user-wide or per-project with bounded durations (capped at 24h) or standing mode, writing records to a user-level SQLite audit DB with 0600 permissions. - Health analysis: The analyze command compares hooks registered in .claude/settings.json and .ai-badger/hooks/hooks.json against observed records, producing findings like never_observed, version_skew, and always_skipped with a health verdict. - Retrieval telemetry: MCP tool-index retrieval events (hit, gate, absent, server_unindexed) are logged with an optional AI_BADGER_DEBUG_REDACT opt-out that drops the query field. - Use Case: After upgrading ai-badger, run analyze --json to confirm every wired hook fires at the new version and no stale plugin-cache copy is still running alongside the scaffold. ## Quick Start Ask the agent to enable ai-badger debug logging for four hours, work normally, then run the analyze command and write up a health report of which hooks fired.
